Boris Johnson during a visit to a school in Somerset. Stefan Rousseau / PA Images Stefan Rousseau / PA Images / PA Images

‘THE ONLY POLL that matters is on election day’.

It’s a maxim used by politicians and weary political watchers but like it or not polling is here to stay. And as anyone following the UK election will know there’s more of it than ever.

But can the polls be trusted?

It’s a question people have been asking more and more after a few years of what’s seemed like unpredictable voting.

The night of the Brexit result started with everyone seemingly convinced that Remain’s lead had held, that was until the hours passed and this was turned upside-down.   

Then there was the shock election of Donald Trump and the seemingly foregone conclusion that Theresa May would win a majority in parliament. 

So is it the case that the polls keep getting it wrong? Well, pollsters say that it’s not quite right to say they missed these events.

In the case of Trump, Hillary Clinton did actually win the lion’s share of the votes, it was only in the swing states that cost her the election that the numbers didn’t stack up. 

With Brexit, polling did show momentum for Leave during the campaign and the eventual result was close enough to be within a margin of error. And even with May in 2017, polling in the weeks leading up to the vote did show that her campaign was going terribly. 

Where the pollsters admit there was a problem was in 2015 UK general election. That race saw an expected hung parliament with Labour perhaps even the largest party.

It actually turned out to be a Conservative majority, a completely incorrect prediction that of course led to some serious questions being asked. 

PA Images / Stefan Rousseau PA Images / Stefan Rousseau / Stefan Rousseau

So to understand how polls can be wrong, it’s best to first look at how they’re supposed to work.  

Polls work on the basis that pollsters get a set of responses from people that’s as representative of the wider electorate as possible – but that often isn’t possible and pollsters then have to make their sample more representative. 

So, for example, if in a poll of 2,000 people the pollsters don’t manage to speak to as many young men as would be representative, a greater weight is given to the young men they did speak to.

There are also lots of other criteria with which pollsters weight their responses, such as how a person voted previously or the likelihood they will vote this time out. 

But one of the problems with opinion polls is that people who have more of an interest in politics are more often likely to participate in them.

Polling companies say they’ve got better at weighting for this but in 2015 one of the big problems they faced was that too many people who told them they would definitely vote were Labour voters.

Lecturer and pollster from Ireland Thinks Dr Kevin Cunningham explains that the voting behaviour of younger people can be among the hardest to predict.

“When you weight it to make sure it’s representative but only have people who are very likely to vote, you tend to over represent certain people,” he says. 

So the worst is perhaps younger voters where the only kinds of people who are 18 to 25, who will say, ‘yeah, I’ll talk about politics and I’ll give you loads of answers and I won’t say undecided’. They tend to be quite liberal or left of centre in particularly and also tend to be like very engaged in politics. And that was a big problem in 2015.

Cunningham points to Yougov as one polling company who overcomes this problem well by weighting respondents based on their political interests rather than how politically engaged they are.  

He says it’s a subtle change that allows the polls to take a better account of people who don’t necessarily follow politics but who are still likely to vote. 

Bums on seats

One of the flaws with polling in the UK is that the first past-the-post system greatly rewards the bigger parties with a disproportionately larger share of seats, while smaller parties get a disproportionately smaller share of seats.

It means that looking at national polls of party support only tells half the picture.  

Getting a more in-depth picture is difficult too because it’s not really feasible for polling companies to do individual polls in each of the UK’s 650 constituencies. 

Instead, what many pollsters use is a system called MRP. Prepare to hear a lot about it this over the coming weeks. 

What MRP stands for is a bit of a mouthful (multilevel regression and post-stratification) but it’s actually quite straightforward to follow.

The model takes a very large polling sample and combines it with Census and other geographical data to produce a more accurate picture of how voters will behave in a given constituency. 

Polling company Survation explains it in more detail on its website but essentially outlines that MRP works by creating demographic types out of respondents to the survey and using Census data to determine how many of each type are in a constituency.

Related Reads

Johnson v Corbyn: 6 key moments from the first televised UK election debate

That is then combined with “additional relevant contextual information to predict how many of these people will vote for each party”. 

The model means that constituency estimates are derived from a national sample, eliminating the problem of not being able to conduct polling in each area.

Cunningham says the model works well and in the last general election it predicted shock Labour wins places like Canterbury and Kensington. 

Kensington you’d probably know instinctively wouldn’t be a Labour seat but Canterbury never had been and it was like a 50/1 shot. But this MRP model predicted they would win in Canterbury and the same sorts of models are now being used for a variety of different issues.

What are the polls saying this time?

But even if we accept that polls can sometimes be misleading, the picture we’re seeing right now appears to be a good one for Boris Johnson. 

The BBC News poll tracker, which takes the average of polling across thirteen different pollsters, puts the Conservative Party at 40% and Labour at 29%. 

It’s a reflection of most of what most of the polling companies are saying and if that gap was replicated in the general election we would instantly be looking at a Conservative majority in parliament.  

UK consultancy Electoral Calculus produces a projected seat analysis based on an average of polls and is currently predicting the Conservatives to win an extra 43 seats, bringing them to 361 and a healthy majority.  

Veteran polling guru Professor John Curtice said similar to Channel 4 News last week, putting the Conservatives’ chances at a majority at 66% and Labour’s chance at the same as being “as close to zero as to be effectively zero”. 

He notes however that Labour doesn’t need to win a majority, but only needs to win enough seats to form a minority government with the support of the SNP and possibly the Liberal Democrats.  

This, Curtice explains, means stopping Johnson getting to the magic number of 326 seats.

Once Boris Johnson is anything more than just a little bit below the 326 mark required for an overall majority, his ability to remain in office becomes increasingly tenuous because there’ll end up being a sufficient group of people who if they want to have a second referendum are going to put a minority Labour administration in.

“The Labour party doesn’t need to get more seats than the Tories, they just simply need to get Boris below 315 definitely it, or below 320.”